job_name = 'debug_shine'
n_gpus = 1
base_params = dict(
model_size='SMALL',
dataset='imagenet',
n_gpus=n_gpus,
n_epochs=100,
seed=42,
restart_from=48,
gradient_correl=False,
gradient_ratio=False,
n_iter=1000,
)
parameters = [
# base_params,
# dict(shine=True, **base_params),
dict(fpn=True, **base_params),
]
executor = get_executor(job_name,
timeout_hour=2,
n_gpus=n_gpus,
project='shine')
jobs = []
with executor.batch():
for param in parameters:
job = executor.submit(train_classifier, **param)
jobs.append(job)
[job.result() for job in jobs]
from mdeq_lib.training_scripts.denoise_train import train_mdeq_denoising
from jean_zay.submitit.general_submissions import get_executor
job_name = 'mdeq_denoise'
executor = get_executor(job_name, timeout_hour=6, n_gpus=1, project='mdeq')
with executor.batch():
for use_bn in [True, False]:
for use_res in [True, False]:
for use_new_residual in [True, False]:
executor.submit(
train_mdeq_denoising,
n_val=20,
use_res=use_res,
use_bn=use_bn,
network_size='SMALL',
use_new_residual=use_new_residual,
grad_clipping=10.,
)
from fastmri_recon.data.scripts.oasis_tf_records_generation import generate_oasis_tf_records
from jean_zay.submitit.general_submissions import get_executor
executor = get_executor('oasis_tfrecords',
timeout_hour=20,
n_gpus=1,
project='fastmri')
with executor.batch():
for mode in ['train', 'val']:
for acq_type in ['radial_stacks', 'spiral_stacks']:
executor.submit(generate_oasis_tf_records,
acq_type=acq_type,
af=4,
mode=mode)
'pdnet': {
'radial_stacks': 'ncpdnet_3d___i6_radial_stacks_mse_dcomp_1612291359',
'spiral_stacks': 'ncpdnet_3d___i6_spiral_stacks_mse_dcomp_1612291359',
},
'unet': {
'radial_stacks': 'vnet_3d___radial_stacks_mse_dcomp_1612291357',
'spiral_stacks': 'vnet_3d___spiral_stacks_mse_dcomp_1612291357',
},
'adj-dcomp': {
'radial_stacks': None,
'spiral_stacks': None,
},
}
executor = get_executor('3dnc_time',
timeout_hour=2,
n_gpus=1,
project='fastmri4')
with executor.batch():
for model, run_ids in model_2_run_ids.items():
for acq_type in ['radial_stacks', 'spiral_stacks']:
executor.submit(
nc_multinet_qualitative_validation,
acq_type=acq_type,
af=4,
model=model,
run_id=run_ids[acq_type],
three_d=True,
refine_smaps=False,
dcomp=True,
normalize_image=False,
n_epochs=8,
from fastmri_recon.models.subclassed_models.denoisers.proposed_params import get_model_specs
from jean_zay.submitit.general_submissions import get_executor
from jean_zay.submitit.fastmri_reproducible_benchmark.mem_fitting_test import test_works_in_xpdnet_train
n_iter_to_try_for_size = {
'medium': range(30, 40),
}
n_primal = 5
job_name = 'xpdnet_tryouts'
executor = get_executor(job_name, timeout_hour=1, n_gpus=1, project='fastmri4')
results = {}
with executor.batch():
for data_consistency_learning in [True, False]:
for model_size_spec, n_iter_to_try in n_iter_to_try_for_size.items():
for model_name, model_size, model_fun, model_kwargs, n_inputs, n_scales, res in get_model_specs(
n_primal):
if model_size != model_size_spec or model_name != 'MWCNN':
continue
for n_iter in n_iter_to_try:
job = executor.submit(
test_works_in_xpdnet_train,
model_fun=model_fun,
model_kwargs=model_kwargs,
n_scales=n_scales,
res=res,
n_iter=n_iter,
multicoil=True,
from fastmri_recon.data.scripts.multicoil_nc_tf_records_generation import generate_multicoil_nc_tf_records
from jean_zay.submitit.general_submissions import get_executor
executor = get_executor('ncmc_tfrecords', timeout_hour=20, n_gpus=1, project='fastmri4')
with executor.batch():
# for mode in ['train', 'val']:
for mode in ['val']:
for acq_type in ['radial', 'spiral']:
executor.submit(generate_multicoil_nc_tf_records, acq_type=acq_type, af=4, mode=mode, brain=True)
from fastmri_recon.evaluate.scripts.nc_eval import evaluate_dcomp
from jean_zay.submitit.general_submissions import get_executor
executor = get_executor('adjoint_dc', timeout_hour=20, n_gpus=1, project='fastmri')
with executor.batch():
for acq_type in ['radial', 'spiral']:
executor.submit(
evaluate_dcomp,
acq_type=acq_type,
af=4,
n_samples=100,
)
from mdeq_lib.training.cls_train import train_classifier
from jean_zay.submitit.general_submissions import get_executor
job_name = 'shine_classifier_cifar_small_fpn'
n_gpus = 4
executor = get_executor(
job_name,
timeout_hour=1,
n_gpus=n_gpus,
project='shine',
torch=True,
no_force_32=True,
)
executor.submit(
train_classifier,
model_size='TINY',
dataset='cifar',
n_gpus=n_gpus,
shine=False,
fpn=True,
n_epochs=25,
)
run_ids = {
4: 'xpdnet_sense__af4_compound_mssim_rf_smb_MWCNNmedium_1606491318',
8: 'xpdnet_sense__af8_compound_mssim_rf_smb_MWCNNmedium_1606491318',
}
model_name = 'MWCNN'
model_size = 'medium'
n_primal = 5
model_specs = list(get_model_specs(force_res=False, n_primal=n_primal))
if model_name is not None:
model_specs = [ms for ms in model_specs if ms[0] == model_name]
if model_size is not None:
model_specs = [ms for ms in model_specs if ms[1] == model_size]
_, model_size, model_fun, kwargs, _, n_scales, res = model_specs[0]
executor = get_executor('postproc_tfrecords', timeout_hour=20, n_gpus=4, project='fastmri')
with executor.batch():
for mode in ['train', 'val']:
for af in [4, 8]:
executor.submit(
generate_postproc_tf_records,
af=af,
mode=mode,
model_fun=model_fun,
model_kwargs=kwargs,
run_id=run_ids[af],
brain=False,
n_epochs=300,
n_iter=10,
res=res,
n_scales=n_scales,
'pdnet': {
'radial': 'ncpdnet_sense___rfs_radial_compound_mssim_dcomp_1611913984',
'spiral': 'ncpdnet_sense___rfs_spiral_compound_mssim_dcomp_1611913984',
},
'unet': {
'radial': 'unet_mc___radial_compound_mssim_dcomp_1611915508',
'spiral': 'unet_mc___spiral_compound_mssim_dcomp_1611915508',
},
'adj-dcomp': {
'radial': None,
'spiral': None,
},
}
executor = get_executor('ncmc_quali',
timeout_hour=2,
n_gpus=1,
project='fastmri4')
with executor.batch():
for model, run_ids in model_2_run_ids.items():
for acq_type in ['radial', 'spiral']:
executor.submit(
nc_multinet_qualitative_validation,
acq_type=acq_type,
af=4,
model=model,
run_id=run_ids[acq_type],
multicoil=True,
refine_smaps=True,
dcomp=True,
normalize_image=False,
contrast='CORPD_FBK',
from fastmri_recon.evaluate.scripts.nc_eval import evaluate_dcomp
from jean_zay.submitit.general_submissions import get_executor
executor = get_executor('adjoint_dc_mc_brain',
timeout_hour=20,
n_gpus=1,
project='fastmri4')
with executor.batch():
for acq_type in ['radial', 'spiral']:
executor.submit(
evaluate_dcomp,
acq_type=acq_type,
af=4,
n_samples=250,
multicoil=True,
brain=True,
)
